In the development of drugs targeted for GSK-3, its selective inhibition is an important requirement owing to the possibility of side effects arising from other kinases for the treatment of diabetes mellitus. A three-dimensional quantitative structure-activity relationship study (3D-QSAR) has been carried out on a set of pyrazolo[3,4-b]pyrid[az]ine derivatives, which includes non-selective and selective GSK-3 inhibitors. The CoMFA models were derived from a training set of 59 molecules. A test set containing 14 molecules (not used in model generation) was used to validate the CoMFA models. The best CoMFA model generated by applying leave-one-out (LOO) cross-validation study gave cross-validation r(cv)(2) and conventional r(conv)(2) values of 0.60 and 0.97, respectively, and r(pred)(2) value of 0.55, which provide the predictive ability of model. The developed models well explain (i) the observed variance in the activity and (ii) structural difference between the selective and non-selective GSK-3 inhibitors. Validation based on the molecular docking has also been carried out to explain the structural differences between the selective and non-selective molecules in the given series of molecules.